Draft version July 5, 2018

Typeset using LATEX twocolumn style in AASTeX62

Extremely Large Extreme-ultraviolet Late Phase Powered by Intense Early Heating in a Non-eruptive Solar Flare
Yu Dai,1, 2, 3 Mingde Ding,1, 2 Weiguo Zong,3 and Kai E. Yang1, 2
1 School of Astronomy and Space Science, Nanjing University, Nanjing 210023, People’s Republic of China
2 Key Laboratory of Modern Astronomy and Astrophysics (Nanjing University), Ministry of Education, Nanjing 210023, People’s Republic
of China
arXiv:1807.01315v1 [astro-ph.SR] 3 Jul 2018

3 Key Laboratory of Space Weather, National Center for Space Weather, China Meteorological Administration, Beijing 100081, People’s

Republic of China

ABSTRACT
We analyze and model an M1.2 non-eruptive solar flare on 2011 September 9. The flare exhibits a
strong late-phase peak of the warm coronal emissions (∼3 MK) in extreme-ultraviolet (EUV), with peak
emission over 1.3 times that of the main flare peak. Multiple flare ribbons are observed, whose evolution
indicates a two-stage energy release process. A non-linear force-free field (NLFFF) extrapolation reveals
the existence of a magnetic null point, a fan-spine structure, and two flux ropes embedded in the fan
dome. Magnetic reconnections involved in the flare are driven by the destabilization and rise of one of
the flux ropes. In the first stage, the fast ascending flux rope drives reconnections at the null point and
the surrounding quasi-separatrix layer (QSL), while in the second stage, reconnection mainly occurs
between the two legs of the field lines stretched by the eventually stopped flux rope. The late-phase
loops are mainly produced by the first-stage QSL reconnection, while the second-stage reconnection is
responsible for the heating of main flaring loops. The first-stage reconnection is believed to be more
powerful, leading to an extremely strong EUV late phase. We find that the delayed occurrence of the
late-phase peak is mainly due to the long cooling process of the long late-phase loops. Using the model
enthalpy-based thermal evolution of loops (EBTEL), we model the EUV emissions from a late-phase
loop. The modeling reveals a peak heating rate of 1.1 erg cm−3 s−1 for the late-phase loop, which is
obviously higher than previous values.

Keywords: Sun: corona — Sun: flares — Sun: UV radiation — Sun: magnetic fields — hydrodynamics

1. INTRODUCTION hard X-ray (HXR) and chromospheric lines (e.g., He II),

Solar flares, one of the most dynamic phenomena in
the solar atmosphere, are manifested as rapid and sig-
nificant enhancements of electromagnetic radiation in a
wide wavelength range. It is widely accepted that so-
lar flares are a result of rapid release of free magnetic
energy stored in the solar corona, and magnetic recon-
nection plays a crucial role in converting the magnetic
energy to plasma heating, particle acceleration, and bulk
mass motions (Parker 1963).
According to the standard two-ribbon solar flare
model, which is conventionally called the CSHKP
model (Carmichael 1964; Sturrock 1966; Hirayama 1974;
Kopp & Pneuman 1976), the evolution of a solar flare
can be divided into two phases: an impulsive phase,
and a following gradual phase. The impulsive phase
is characterized by a rapid increase of the emissions in
Corresponding author: Yu Dai
ydai@nju.edu.cn
indicating a prompt response of the solar lower atmo-
sphere to nonthermal electron bombardment and/or
thermal conduction caused by the initial energy re-
lease. Since the energy transported downward cannot
be effectively radiated away at the solar lower atmo-
sphere (Antiochos & Sturrock 1976), a hot evapora-
tive flow is consequently driven into the flare loops
(Antiochos & Sturrock 1978), which then brighten up
in soft X-ray (SXR) and coronal extreme-ultraviolet
(EUV) lines. As the flare plasma cools down, these
emissions peak sequentially in an order of decreas-
ing temperatures, and then decay gradually toward
the background levels, constituting the gradual phase
(Chamberlin et al. 2012).
The original CSHKP model and its variants are a
two-dimensional (2D) model in nature. Nevertheless,
a real solar flare takes place in a three-dimensional (3D)
magnetic field, making its evolution fairly more com-
plicated than that expected based on a simplified 2D
model. By using EUV irradiance observations with the
EUV Variability Experiment (EVE; Woods et al. 2012)
2 Dai et al.
on board the recently launched Solar Dynamics Obser-
vatory (SDO ; Pesnell et al. 2012) mission, Woods et al.
(2011) discovered a new phenomenon in some flares,
namely, an “EUV late phase”. Observationally, the
EUV late phase is seen as a second peak in the warm
coronal emissions (∼3 MK) several tens of minutes to a
few hours after the GOES SXR peak. There are, how-
ever, no significant enhancements of the SXR or hot
coronal emissions (∼ 10 MK or higher) in the EUV late
phase, and spatially resolved imaging observations such
as those from the Atmospheric Imaging Assembly (AIA;
Lemen et al. 2012) also on board SDO reveal that the
secondary late-phase emission comes from another set
of flare loops higher and longer than the main flaring
loops.
In a preliminary statistics of 25 EUV late-phase solar
flares occurring during the first year of SDO normal op-
erations, Woods et al. (2011) found that about half of
them took place in two active regions (ARs), implying a
specific magnetic configuration of the ARs in which EUV
late-phase flares are preferentially produced. In case
studies, photospheric magnetograms of the flare-hosting
ARs reveal that the EUV late-phase flares are commonly
involved in a multipolar magnetic field, which exhibits
either a symmetric or asymmetric quadrupolar configu-
ration (Hock et al. 2012; Liu et al. 2013), or a parasitic
polarity embedded in a large-scale bipolar magnetic field
(Dai et al. 2013; Sun et al. 2013; Masson et al. 2017).
Such a magnetic configuration facilitates the existence
of two sets of loops that are magnetically related but
distinct in length, as further confirmed by the force-free
coronal magnetic field extrapolations (Sun et al. 2013;
Li et al. 2014; Masson et al. 2017).
The EUV irradiance output from the Sun can drive
disturbances in Earth’s ionosphere and thermosphere,
particularly during solar flares (Kane & Donnelly 1971).
Hence, the origin of EUV late phase in a solar flare is of
interest for its potential geo-effectiveness. Since many
theoretical works on the cooling of flare plasmas had
shown that the cooling time of a flare loop increases as
the loop length increases (e.g., Cargill et al. 1995), it was
naturally proposed that the delayed occurrence of an
EUV late phase is the result of a long-lasting cooling pro-
cess in the long late-phase loops initially heated simul-
taneously with the main flaring loops (Liu et al. 2013;
Masson et al. 2017). In this sense, the late phase is actu-
ally the gradual phase in long flare loops. Nevertheless,
there was also accumulating evidence for the existence of
a delayed secondary heating, which might be responsible
for the production of the EUV late phase (Hock et al.
2012; Dai et al. 2013). In fact, the two mechanisms are
not mutually exclusive, and in some EUV late-phase
flares they may both play a role (Sun et al. 2013).
By using the model called enthalpy-based thermal
evolution of loops (EBTEL; Klimchuk et al. 2008;
Cargill et al. 2012; Barnes et al. 2016), Li et al. (2014)
and Dai & Ding (2018) numerically probed the produc-
tion of EUV late-phase flares. In particular, Dai & Ding
(2018) found that even with an equal energy partition
between the late-phase loop and main flaring loop, the
warm coronal late-phase peak is still significantly lower
than the corresponding main flare peak. This result may
offer a clue to why EUV late-phase flares are rare among
all solar flares. Nevertheless, as shown in the statistics
in Woods et al. (2011), the ratio of the late-phase peak
to main flare peak in their sample can be as high as 4.
Liu et al. (2015) have recently reported a long-duration
but non-eruptive flare in which the late-phase peak is
∼ 2.1 times higher than the main flare peak. For this
reason they referred to it as an “extremely large EUV
late phase”. By analyzing the kinematic and thermody-
namic evolution of the flare, Liu et al. (2015) proposed
that an “erupted-but-failed” hot structure, which is
most presumably a magnetic flux rope, serves as a per-
sistent heating agent for this extremely large EUV late
phase. Wang et al. (2016) argued that a strong con-
straint of the overlying arcades may prevent a flux rope
from escaping, and cause the energy carried by the flux
rope to be re-deposited into the thermal emissions that
form a stronger late phase. This argument was fur-
ther validated through a comparative study of 12 EUV
late-phase flares (Wang et al. 2016) showing that the
relative late-phase peaks of the non-eruptive flares are
systematically stronger than those of the eruptive flares.
In addition to the persistent heating scenario proposed
by Liu et al. (2015), we are wondering if an impulsive
heating on the late-phase loops, which is temporally
close to the main flare heating, can also produce a signif-
icantly large EUV late phase. In this paper, we report
observations of another non-eruptive solar flare that ex-
hibits an extremely large EUV late phase. The detailed
data analysis shows that this large EUV late phase is
mainly powered by an intense heating even earlier than
the main flare heating. The paper is organized as fol-
lows. In Section 2, we describe the evolution of the flare.
In Section 3, we explore the magnetic topology of the
flare-hosting AR based on a non-linear force-free field
(NLFFF) extrapolation, and in Section 4, we use the
EBTEL model to synthesize the EUV emissions from a
late-phase loop. The results are discussed in Section 5,
and a brief summary is presented in Section 6.
2. EVOLUTION OF THE FLARE
The flare-hosting AR NOAA 11283 has been ex-
tensively studied in the literature (e.g., Feng et al.
2013; Jiang et al. 2014; Liu et al. 2014; Xu et al. 2014;
Romano et al. 2015; Ruan et al. 2015; Zhang et al.
2015). Its passage through the visible disk from 2011
August 31 to September 11 can be divided into two
periods of roughly equal durations. During the first
period, the AR was simply bipolar, indicative of a low
flare level. As revealed by the GOES flare catalog, the
AR had just harbored eight flares of classes at most
C2.2 during this period. From September 4 to 5, a

GOES Flux (W m−2) (a) 1−8 Å
10−5
10−6
11:30 12:00 12:30
40
13:00
(b) EVE 304 Å
30 He II log(T/K)=4.7
20
10
0 very prompt rise followed by a less impulsive one. After
−10
3 (c) EVE 133 Å
2
Fe XX/XXIII log(T/K)=7.0
1
EVE Irradiance (µW m−2)
3 8
(d) EVE 94 Å
2 Fe XVIII log(T/K)=6.8
1
0
6 (e) EVE 335 Å
Fe XVI log(T/K)=6.5
4
2
0
(f) EVE 171 Å
2 Fe IX log(T/K)=5.9
1 20
0
−1
11:30 12:00 12:30 13:00 13:30 14:00
Time (UT on 2011−Sep−09)
Figure 1. Time profiles of the GOES 1–8 Å SXR flux (a),
and the EVE full-disk irradiance (black) and AIA sub-region
intensities (colored) in several lines and passbands (b)–(f)
for the 2011 September 9 M1.2 flare. The vertical dashed
lines mark the peak times of the GOES flux and AIA in-
tensities. In deriving the EVE profiles, a line-dependent
smoothing boxcar ranging from 60 to 300 s is applied to
the original data points (plus signs) to enhance the ratio of
signal to noise. The region over which AIA pixel count rates
are summed covers a field-of-view (FOV) of 200′′ ×200′′ en-
closing the flare-hosting AR. Note that the EVE and AIA
profiles are plotted with the corresponding background levels
subtracted. For clarity, in each panel the background levels
for the EVE and AIA profiles are plotted (horizontal dashed
lines) with a vertical offset.
persistent emergence of positive magnetic flux inside
the leading negative polarity of the AR was observed,
which not only complicated the magnetic topology of
the AR, but also enhanced its flare-productivity. As a
result, a series of much more energetic flares including
two X-class and five M-class ones were produced in the
AR during the second period, and furthermore, some of
them exhibited an evident EUV late phase. The pro-
duction of EUV late phase in the X2.1 eruptive flare on
September 6 has been investigated by Dai et al. (2013).
In this work, we focus on an M1.2 non-eruptive flare on
September 9 when the AR had been decaying.
EUV LATE PHASE IN A NON-ERUPTIVE FLARE 3
2.1. Extremely Large EUV Late Phase
GOES Class
M
Figure 1(a) shows the GOES 1–8 Å light curve of
C
the flare under study. Following a small bump around
12:30 UT, the SXR flux starts to rise rapidly from
13:30 14:00 14:30
15 ∼12:39 UT and reaches its peak at 12:49:19 UT (out-
AIA 304 Å
10
lined by the vertical dashed line). It is clearly seen that
the rise phase of SXR exhibits a two-stage evolution: a
5
0
30 the peak, the SXR emission also reveals two distinct de-
AIA 131 Å
20
cay periods: a rapid decay, and a following much longer
and less sloped one lasting until ∼13:40 UT when the
10
SXR emission is elevated again.
AIA Intensity (106 DN s−1)
0
To further study the thermal evolution of the flare,
AIA 94 Å
6 we plot in Figures 1(b)–(f) the time profiles of the
4 background-subtracted irradiance in several EVE spec-
2 tral lines (we adopt a 5-minute average between 11:30
0 and 11:35 UT as the pre-flare background). EVE mea-
AIA 335 Å 4
3
sures full-disk integrated solar EUV irradiance from 1
2 to 1050 Å with 1 Å spectral resolution and 10 s time
1 cadence. By integrating EVE spectra (EVS files) over
0 specified spectral windows, the irradiance of some “iso-
AIA 171 Å 40
30
lated” EUV lines (EVL files) can be derived. Here we
choose five lines from the Multiple EUV Grating Spec-
10 trographs (MEGS)-A component of EVE, which covers
0
a wavelength range of 65–370 Å with a nearly 100% duty
14:30
cycle.
Several emission patterns of the flare can be seen from
the EVE profiles. First, the cold chromospheric EVE
304 Å emission (dominated by the He II λ303.8 Å line,
log(T /K) ∼ 4.7, Figure 1(b)) shows a very impulsive rise
that quickly reaches its peak at 12:42:54 UT, character-
izing the flare’s impulsive phase. In spite of a high noise
level in the original data, the smoothed profile (after
applying a 300 s smoothing boxcar) of the cool coronal
EVE 171 Å emission (Fe IX λ171.1 Å, log(T /K) ∼ 5.9,
Figure 1(f)) also reveals an obvious rise with a local
maximum near the peak time in EVE 304 Å. This im-
plies that the chromosphere and transition region (TR),
as responding to the initial energy release, are instantly
heated to at least ∼ 1 MK (Chamberlin et al. 2012;
Milligan et al. 2012).
Second, the time profile of the hot coronal EVE 133 Å
emission (blended by Fe XX λ132.8 Å and Fe XXIII
λ132.9 Å, log(T /K) ∼ 7, Figure 1(c)), closely resem-
bling the GOES SXR light curve thanks to their similar
temperature sensitivity, also exhibits a dual-rise followed
by a dual-decay, and peaks nearly simultaneously with
SXR at 12:49:24 UT. It is noted that the first-stage
rise of the EVE 133 Å/GOES SXR emission is tempo-
rally correlated to the impulsive rise of the EVE 304 Å
emission. After the EVE 133 Å peak, the cooler coro-
nal emissions peak sequentially at 12:51:34 UT in EVE
94 Å (Fe XVIII λ93.9 Å, log(T /K) ∼ 6.8, Figure 1(d))
and at 12:54:44 UT in EVE 335 Å (Fe XVI λ335.4 Å,
log(T /K) ∼ 6.5, Figure 1(e)). The small time lags be-
tween these emission peaks indicate a fast cooling pro-
4 Dai et al.
cess in the corresponding flare plasma, whose inferred
cooling rate (∼ 2 × 104 K s−1 ) is comparable to those
found in Chamberlin et al. (2012).
In addition to the first peaks, there exist second con-
spicuous emission peaks in both EVE 94 Å and EVE
335 Å, which occur at 13:05:54 and 13:37:34 UT, respec-
tively. The time lag between the second and first peaks
in EVE 335 Å is 43 minutes, lying close to the lower
limit of the time delay of a warm coronal late-phase peak
with respect to the corresponding main flare peak (41
to 204 minutes) in the sample of Woods et al. (2011).
The emission ratio between the two peaks is 1.54 ± 0.20
(considering the fluctuations in the original data), nearly
twice the average value (0.8) in Woods et al. (2011).
Based on this high peak emission ratio, we adopt the
same definition by Liu et al. (2015) and hence name this
second peak an extremely large EUV late-phase peak.
It is interesting that a large late phase is also evident
in EVE 171 Å, whose peak occurs further later (after
14:00 UT).
2.2. Correspondence in Spatially Resolved
Observations
To resolve activities responsible for the flare emissions,
we inspect AIA images of the flare-hosting AR. AIA
presents multiple simultaneous full-disk images of the
TR and corona in 10 passbands with a pixel size of 0.′′ 6
and a temporal resolution of 12 s (24 s) for the EUV
(UV) passbands. Here we choose a sub-region that cov-
ers a field of view (FOV) of 200′′ ×200′′ (x ∈ [610′′ , 810′′ ],
y ∈ [80′′ , 280′′]) enclosing the entire AR. The online An-
imation 1 presents the flare evolution in six coronal pass-
bands of AIA. Since the evolution behavior of the flare
in the AIA 211 and 193 Å passbands is generally simi-
lar to that in AIA 171 Å, we only extract characteristic
snapshots in AIA 131 (Fe XXI, Fe VIII), 94 (Fe XVIII,
Fe X), 335 (Fe XVI), and 171 Å (Fe IX) from the anima-
tion, and display them in Figure 2. Note that the dom-
inant ion of each AIA passband is either the same as or
of a similar formation temperature to that of the corre-
sponding EVE line selected in Figure 1 (O’Dwyer et al.
2010). As indicated by the arrows in Figure 2(d1), there
are two low-lying filaments located in the pre-flare AR:
one north-south oriented (hereafter NS), and the other
east-west oriented (hereafter EW). The filaments are ini-
tially most discernible in the cool passbands such as AIA
171 Å (and also AIA 131 Å because of its low temper-
ature sensitivity around 0.6 MK in addition to the high
temperature response peak over 10 MK). It seems that
the southern end of the NS filament is connected to the
eastern end of the EW filament, apparently constituting
a single L-shaped filament. Nevertheless, the following
flare evolution shows that these two filaments are indeed
evolving separately.
The precursor of the flare is seen as a pre-flare bright-
ening in all AIA passbands around 12:30 UT (the first
row of Figure 2), which should be responsible for the
co-temporal small bumps seen in the GOES and EVE
time profiles in Figure 1. The brightening first appears
in a set of sheared arcades enveloping the EW filament
(highlighted by the arrow in Figure 2(c1)), and then
shows a quick eastward propagation along the EW fila-
ment. When the pre-flare brightening propagates to the
NS filament, this filament is destabilized and starts to
ascend rapidly. As the filament rises up, its body ab-
sorbs more background coronal emission along the line-
of-sight (LOS). As a result, the ascending filament be-
comes much more clearly seen as a curved dark feature
in AIA 335 Å (indicated by the yellow arrow in Fig-
ure 2(c2)). Meanwhile, multiple flare ribbons brighten
up quickly in all AIA passbands, coinciding with the
flare’s impulsive phase (the second row of Figure 2). As
pointed out by the white arrows in Figure 2(c2), the flare
ribbons not only develop from two footpoints of the as-
cending filament, but also appear in some other remote
locations to the east and north of the filament. Near
the peak of the impulsive phase (∼12:42 UT), several
long brightening loops anchored on the eastern remote
flare ribbon are already observable in the hot passbands
such as AIA 131 (Figure 2(a2)) and 94 Å (Figure 2(b2)),
whereas they are invisible in the cooler AIA 335 (Fig-
ure 2(c2)) and 171 Å (Figure 2(d2)) passbands.
The fast rising filament then slows down and finally
stops at a certain altitude. Material of the filament falls
down towards its two footpoints along the spine, making
the filament eventually drained off (evident in Anima-
tion 1). Coronagraph images from the Large Angle and
Spectrometric Coronagraph (LASCO; Brueckner et al.
1995) on board the Solar and Heliospheric Observatory
(SOHO ) spacecraft have not recorded any coronal mass
ejections (CMEs) during this period, further confirming
that the flare is non-eruptive. Around the peak time
of GOES SXR emission (the third row of Figure 2), in-
tensely brightening loops appear underneath the previ-
ously ascending filament (outlined by the arrow in Fig-
ure 2(c3)). In the framework of the CSHKP model, these
loops correspond to the post-flare loops in the wake of
an erupting flux rope, which exhibit the main phase of
the flare. Note that in this row, each frame is taken at
the time when the corresponding sub-region intensity at-
tains a local maximum. Hence the time lags between the
frames in the same row should imply a cooling process
in the main flaring loops. In addition, the long brighten-
ing loops anchored on the remote flare ribbon, which can
already be seen in the impulsive phase, are now fully de-
veloped as revealed in AIA 131 Å (Figure 2(a3)). Com-
pared to the main flaring loops, these longer brightening
loops are less bright but cover a much larger spatial ex-
tent. Nevertheless, they are still invisible in AIA 335
(Figure 2(c3)) and 171 Å (Figure 2(d3)).
The main flaring loops quickly fade out, while the long
brightening loops continues to brighten up and reaches
an emission peak sequentially in cooler AIA passbands,
as shown in the bottom row of Figure 2. Obviously,
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE 5

0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.5 1.0 1.5 2.0 2.5 3.0 2.0 2.5 3.0 3.5
log(IAIA/DN s−1) log(IAIA/DN s−1) log(IAIA/DN s−1) log(IAIA/DN s−1)
(a1) AIA 131 Å (b1) AIA 94 Å (c1) AIA 335 Å (d1) AIA 171 Å

Pre−Flare Brightening
250
Pre−Flare
Brightening
Y (arcsec)
Filaments
200

150

100 12:29:57 UT 12:29:50 UT 12:29:51 UT 12:29:50 UT

(a2) (b2) (c2) Flare Ribbons (d2)

250

Impulsive Phase
Ascending
Y (arcsec)

Filament
200

150

100 12:42:21 UT 12:42:14 UT 12:42:15 UT 12:42:14 UT

(a3) (b3) (c3) (d3)

250
Main Flaring Loops

Main Phase
Y (arcsec)

200

150

100 12:50:21 UT 12:51:38 UT 12:54:03 UT 12:57:26 UT

(a4) (b4) (c4) Late−Phase Loops (d4)

250

Late Phase
Y (arcsec)

200

150

100 14:01:21 UT 13:05:02 UT 13:37:39 UT 14:01:12 UT

650 700 750 800 650 700 750 800 650 700 750 800 650 700 750 800
X (arcsec) X (arcsec) X (arcsec) X (arcsec)

Figure 2. Snapshots of the flare evolution in four AIA coronal passbands. The frames are extracted from the online animation
to characterize different phases of the flare evolution, and the arrows highlight some characteristic structures, which are decribed
in detail in the text.
the long brightening loops in AIA 335 Å (indicated by low temperature (∼0.6 MK) to which the AIA 131 Å
the arrows in Figure 2(c4)) are responsible for the ex- passband is also sensitive.
tremely large EUV late phase revealed in EVE 335 Å. We further calculate the sub-region intensities by sum-
Hence we are convinced to identify them as late-phase ming the count rates over all AIA pixels in the FOV
loops. Furthermore, the late-phase loops in AIA 335 Å of Figure 2. The intensity profiles (after subtracting a
(∼13:37 UT, Figure 2(c4)) are morphologically similar pre-flare background averaged from 11:30 to 11:35 UT,
to those seen earlier in the hotter passbands like AIA the same as that chosen in generating the EVE profiles)
131 (∼12:50 UT, Figure 2(a3)) and 94 Å (∼13:05 UT, in five AIA passbands (here we include AIA 304 Å in
Figure 2(b4)), and later in the cooler passband like AIA addition to the four passbands displayed in Figure 2)
171 Å (∼14:01 UT Figure 2(d4)), indicating that they are accordingly overplotted in Figures 1(b)–(f). In each
are the same structures seen in different passbands and panel, the evolution of the AIA sub-region profile is sim-
also at different times. Here the larger time lags be- ilar to that of the EVE profile, with the AIA peaks (out-
tween the appearance times of loops in different AIA lined by the vertical dashed lines) occurring nearly si-
passbands reflect a much slower cooling process in the multaneously with the corresponding EVE peaks (typ-
long late-phase loops than that in the main flaring loops. ically within 1 minute). In particular for the pairs of
As mentioned above, the re-appearance of the late-phase EVE 133 Å/AIA 131 Å, EVE 94Å/AIA 94 Å, and EVE
loops in AIA 131 Å around 14:01 UT (Figure 2(a4)) 335 Å/AIA 335 Å, the Pearson correlation coefficient
means that the late-phase loops have cooled down to a between the two paired profiles is as high as 0.98–0.99.
Moreover, the emission ratio of the late-phase peak to
6 Dai et al.
the main flare peak in AIA 335 Å is 1.36, which is, if
considering the commonly broader temperature cover-
age of an AIA passband compared to that of the corre-
sponding EVE line, consistent with the ratio revealed in
EVE 335 Å. We have also changed the size of the sub-
region used for intensity calculation. It is found that
as long as the size of the sub-region is large enough to
contain the whole AR, its variation only marginally af-
fects the resulting AIA profiles. All these factors suggest
that during this period, the activities taking place inside
the AR contribute predominantly to the variabilities in
full-disk integrated solar emissions in EUV wavelengths,
especially for those formed in hot and warm coronal plas-
mas. Since the Fe IX emission mainly comes from the
cool bulk corona, emission fluctuations outside the AR
may significantly influence the overall irradiance in EVE
171 Å, hence making the original data for this line rather
noisy, as opposed to those in other EVE lines. Finally,
it is worth noting that there are a small spike around
13:15 UT and a moderate bump after 13:40 UT in the
GOES and EVE profiles, which nevertheless have no
counterparts in the AIA sub-region profiles. By care-
fully checking the AIA full-disk images, it is found that
they are caused by two small brightening events occur-
ring in another AR NOAA 11289.
2.3. Two-Stage Energy Release
The evolution of the flare is also traced in the chro-
mospheric AIA 1600 Å passband, as demonstrated in
the online Animation 2. Emission in this passband in-
cludes the continuum formed around the temperature
minimum region and the C IV λ1548.2/1550.8 Å dou-
blet lines formed in the upper chromosphere and TR.
During a solar flare, the optically thin C IV lines are
significantly enhanced, showing a prompt and sensitive
response to the flare energy release. Therefore, the evo-
lution in AIA 1600 Å can be regarded as a proxy for
flare heating (Li et al. 2012; Qiu et al. 2012; Zhu et al.
2018). Some snapshots of the animation, as well as an
LOS magnetogram of the AR taken near the time of
flare by the Helioseismic and Magnetic Imager (HMI;
Scherrer et al. 2012) on board SDO, are displayed in
Figures 3(a)–(d). Three main magnetic polarities are
identified and labeled as P1, N1, and P2, of which P2 is
a parasitic positive polarity embedded in the host neg-
ative polarity N1 (Figure 3(a)). During the course of
the flare, flare ribbons are observed in all of these po-
larities (Figures 3(b)–(d)), like those seen in Figure 2
but free of any possible contaminations from the loop
structures. The appearance of multiple flare ribbons is
believed to be a natural consequence of the multipolar
magnetic field. When overplotting the contours of LOS
magnetic field onto the AIA 1600 Å image (Figure 3(c)),
it is seen that the flare ribbons generally occur in loca-
tions of magnetic concentration.
The AIA 1600 Å intensity profile of the AR (purple)
shown in Figure 3(e) reveals two peaks at 12:42:41 and
12:47:53 UT (highlighted by the vertical dashed lines),
respectively, indicative of two main episodes of energy
release taking place in quick succession. With the aid
of the profile, we select the AIA 1600 Å snapshots close
to or at the times of the intensity peaks, as displayed in
Figures 3(b)–(d). The first-stage energy release is very
energetic; flare ribbons brighten up promptly, coincid-
ing with the flare’s impulsive phase. Around the peak of
the ribbon brightening (Figures 3(b) and (c)), the flare
ribbons are manifested as an elongated one (R1) located
in the eastern positive polarity P1 and a semi-circular
one (R2) located in the western negative polarity N1.
R1 corresponds to the remote flare ribbon seen in Fig-
ure 2, whereas R2 is not continuous but broken at its
middle into two segments: R2n in the north and R2s in
the south. A third flare ribbon is believed to exist in
the parasitic positive polarity P2. However, saturation
of the CCD pixels in this region largely masks the rather
compact ribbon, making it undistinguishable from the
nearby ribbon segment R2s. To validate our specula-
tion, we reconstruct the Reuven Ramaty High Energy
Solar Spectroscopic Imager (RHESSI ; Lin et al. 2002)
image at the energy band of 12–25 keV in the interval of
12:41:54–12:42:18 UT, which reveals two HXR sources
(outlined by the red contours in Figure 3(b)) located in
the “masked” compact ribbon and the southern ribbon
segment R2s, respectively. When overplotting the HXR
sources onto the simultaneous AIA 335 Å image inserted
in Figure 3(b), it is further found that they are excel-
lently co-spatial with the two footpoints of the ascending
NS filament.
Compared to the first-stage energy release, the second-
stage energy release is more gentle; the flare ribbons
re-brighten up moderately after the first peak. For the
eastern remote ribbon R1, the brightening is observed
to propagate from north to south along the ribbon. For
the western semi-circular ribbon R2, a newly brighten-
ing ribbon segment R2m appears and becomes the most
prominent section along this ribbon, filling the gap be-
tween the previously disconnected ribbon segments R2n
and R2s. Meanwhile, the previously masked compact
ribbon in the parasitic polarity P2 now becomes clearly
discernible, which is identified as R3 (Figure 3(d)). As
the energy release goes on, both R2m and R3 experience
a separation motion away from the polarity inversion
line (PIL) between them, which is usually observed in a
two-ribbon solar flare. All these evolution patterns are
clearly revealed in Animation 2.
In addition to the time profile of the full AR, in Fig-
ure 3(e) we also plot the time profiles of the AIA 1600 Å
intensity for regions covering the eastern and western
ribbons, which are outlined in Figure 3(c) by the orange
and blue boxes, respectively. During the first-stage en-
ergy release, the intensities in both regions show a syn-
chronous enhancement, with the first peaks in all time
profiles occurring at exactly the same time. Neverthe-
less, the second intensity peak for the region of eastern
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE 7

−1000 −500 0 500 1000 1.5 2.0 2.5 3.0 3.5

BLOS (G) log(IAIA/DN s−1)
AIA 335 Å 14
(a) HMI LOS (b) AIA 1600 Å 12:42:03 UT (e) 1600 Å Full AR
12 Western Ribbons

AIA Intensity (106 DN s−1)

250 S0
Eastern Ribbon
10
R2n
Y (arcsec)

200 P2 8
P1 R1
6
150 N1 4
R2s
RHESSI 12−25 keV
12:41:54−12:42:18 UT 2
100 12:20:07 UT 12:42:17 UT 0

GOES Flux Deriv. (10−8 W m−2 s−1)

10 3.0

RHESSI Count Rate (103 c s−1)

(c) AIA 1600 Å (d) AIA 1600 Å (f) GOES 1−8 Å
250 8 RHESSI 12−25 keV
2.5
6
2.0
Y (arcsec)

200 S2
R3 4
S1 1.5
S3
2
150 R2m
1.0
0
0.5
100 12:42:41 UT −2
12:47:53 UT
0.0
650 700 750 800 650 700 750 800 12:30 12:40 12:50 13:00 13:10
X (arcsec) X (arcsec) Time (UT on 2011−Sep−09)

Figure 3. Left and middle: HMI line-of-sight (LOS) magnetogram of AR 11283 taken at 12:20 UT (a), and snapshots showing
the flare evolution in AIA 1600 Å (b)–(d), which are extracted from the online animation. The LOS magnetogram reveals
three main magnetic polarities, whose contours are overpoltted in panel (c), with levels of ±200, ±600, and ±1000 G (pink for
positive, and cyan for negative). Multiple flare ribbons (prefixed with the letter “R”) are identified in the AIA 1600 Å images.
The inset in panel (b) is a simultaneous AIA 335 Å image, whose FOV is outlined by the dashed box. Also overplotted in this
panel is a simultaneous RHESSI image at energies of 12–25 keV reconstructed with the Pixon algorithm, with contour levels
corresponding to 30%, 50%, 70%, and 90% of the maximum intensity, respectively. Four slices (S0–S3) are drawn as the yellow
dashed lines or curves in panels (b) and (d), for which the time-distance plots are presented in Figure 4. Right: time profile of
the AIA 1600 Å intensity for different regions (e) and time profiles of soft (derivative) and hard X-rays (f). The vertical dashed
lines indicate the peak times for the corresponding profiles. The AIA intensity is integrated over sub-regions enclosed by the
colored boxes in panel (c).
ribbon is much less discernible than that for the region As its ascending speed gradually approaches zero, the
of western ribbons. It implies that during the second- filament disappears abruptly in the S0 plot, which is
stage energy release, the majority of the released energy due to the falling of material from the filament apex.
should be deposited into the loops anchored on the west- Shortly after the “disappearance” of the filament
ern ribbons. (within 1 minute), the second-stage energy release comes
We use time–distance plots to further study the en- into play. Brightening occurs successively from north
ergy release process. Four slices are selected and drawn to south along the northern section of R1, revealing
in Figures 3(b) and (d), among which S0 extends across an apparent motion speed of 180 km s−1 (Figure 4(b)).
the apex of the ascending filament while S1–S3 lie either Elongation motions of flare ribbon brightening along the
parallelly or perpendicularly to the flare ribbons. The PIL have been both theoretically modeled and obser-
temporal evolution along these slices in AIA 335 Å (S0) vationally seen in solar flares (e.g., Priest & Longcope
and 1600 Å (S1–S3) is shown in Figure 4. A close cor- 2017; Qiu et al. 2017). In this case, the observed slip-
relation between the rise of the filament and the bright- ping of the ribbon brightening is more likely linked to
ening of the flare ribbons is found. A linear fit to the 3D slipping/slip-running reconnection (Aulanier et al.
trajectory of the ascending filament apex reveals a rise 2006, 2007) taking place in a quasi-separatrix layer
speed of 175 km s−1 projected onto the plane of the (QSL; Démoulin et al. 1996). In addition to the appar-
sky (Figure 4(a)). Coinciding with the fast rise of the ent elongation motion of R1, both the newly brightening
filament, the flare ribbons (R1, R2n, R2s, and possi- ribbon segment R2m (Figure 4(c)) and the previously
bly masked R3) brighten up simultaneously and quickly masked ribbon R3 show a separation motion from the
reach the first peaks exactly at the time when the fast as- PIL (Figure 4(d)). Compared to R3, the motion of R2m
cending filament turns to decelerate (Figures 4(b)–(d)). is more reliable to trace. A linear fit to the outer bound-
8 Dai et al.
60
(a) S0 AIA 335 Å NW light curve is that the satellite has orbited into the Earth
Distance (arcsec) 50
shade at this time.

km s −1
40
Ascending
30 Filament

175
20
3. MAGNETIC TOPOLOGY OF THE AR
10
0 SE
Magnetic topology is crucial for understanding the
12:30 12:35 12:40 12:45 12:50 12:55 13:00
Peak 1 Peak 2 magnetic reconnection process in a solar flare and the
100 (b) S1 AIA 1600 Å N consequent morphological evolution of the flare loops
and ribbons. In a typically low β (the ratio of gas pres-

18
80

0k
sure to magnetic pressure) corona, we explore the mag-

m
60

s
R1 netic topology of the flare-hosting AR with the aid of

−1
40

20
an NLFFF extrapolation.
0 S We adopt the optimization method proposed by
120 (c) S2 AIA 1600 Å N Wheatland et al. (2000) and implemented by Wiegelmann
(2004) to perform the NLFFF extrapolation. To pre-
Distance (arcsec)

100
R2n
80
60
R2s
40
20
0 S
30
(d) S3 AIA 1600 Å
25
20
15
10
5
0 R2m
12:30 12:35 12:40
Time (UT on 2011−Sep−09)
Figure 4. Time-distance plots along the selected slices in
AIA 335 Å (a) and 1600 Å (b)–(d), within which some char-
acteristic structures are identified and labeled. Linear fit-
tings are applied to some moving structures traced by the
yellow dotted lines, with the fitted speeds labeled along these
lines. Overplotted in panels (a) and (d) is the AIA 1600 Å
intensity profile of the AR (the same as that plotted in Fig-
ure 3(e)) revealing two peaks by the vertical dashed lines in
all panels.
ary of R2m reveals a separation speed of 20 km s−1 . To
compensate for the projection shortening at the flare
position of ∼N13◦ W52◦ , we multiply the fitted speed
by a factor of ∼1.6, and obtain a corrected separation
speed of ∼30 km s−1 , which is consistent with the results
previously found in two-ribbon flares (e.g., Asai et al.
2004; Miklenic et al. 2007).
The two episodes of energy release are also evident
in X-rays. Figure 3(f) shows the derivative of the
GOES 1–8 Å flux (black), which is generally taken as
a proxy for the HXR emission owing to the “Neupert
effect” (Neupert 1968). Two peaks are readily resolved
at 12:42:33 and 12:47:45 UT (indicated by the vertical
dashed lines), respectively, in perfect coincidence with
the corresponding AIA 1600 Å peaks (within 10 s). The
overplotted RHESSI 12–25 keV light curve (red) reveals
only one peak at 12:42:26 UT, which coincides with the
first peak in the GOES derivative curve. The reason
for the absence of a second peak in the RHESSI HXR
R2m
pare the bottom boundary, we use an HMI vector mag-
netogram of the AR taken close to the time of flare
occurrence (∼11:58 UT), which is selected from the
Space-weather HMI Active Region Patches (SHARPs).
W The pipeline to generate SHARP data products includes
R3
the removal of the 180◦ ambiguity in transverse field
(Metcalf et al. 2006; Leka et al. 2009), and the remap-
20 ping of the magnetic field vector (Gary & Hagyard 1990)
km
s −1 in a cylindrical equal-area (CEA; Calabretta & Greisen
E 2002) projection (for more details, see Bobra et al. 2014
12:45 12:50 12:55 13:00 and Hoeksema et al. 2014). The grid size used for the
SHARP sampling is 0.03◦ in heliocentric angle, corre-
sponding to a pixel size of ∼0.36 Mm. To save the
computation time while not significantly losing the es-
sential features in the magnetic field modeling, we ex-
tract a sub-region that is large enough to cover all main
polarities in the original SHARP magnetogram, and
degrade the sampling resolution by a factor of 2. A
further preprocessing is applied to the area of interest
to minimize the net magnetic force and torque on the
boundary (Wiegelmann et al. 2006). The extrapolation
is then conducted in a box of 300 × 200 × 200 grid points
uniformly spaced in the computation domain.
The magnetic topology of the AR based on our
NLFFF extrapolation is presented in Figure 5. The
most prominent features are a 3D magnetic null point
located at a height of 11.8 Mm, and inner (green)
and outer (blue) fan-spine magnetic field lines pass-
ing around it (Lau & Finn 1990). With the aid of the
normal component (Bz ) of the SHARP vector magne-
togram (Figures 5(a)–(c)), it is seen that the footpoints
of the inner and outer spine field lines reside in the
parasitic and eastern positive polarities P2 and P1,
respectively, while the fan field lines form a dome-like
structure, whose base intersects with the solar surface in
the negative polarity N1. Embedded in the fan dome,
two magnetic flux ropes (red for the NS-oriented one
and yellow for the EW-oriented one) are found to ex-
ist. Through co-alignment with the AIA coronal images
(not shown here), it is found that the flux ropes are ap-
proximately co-spatial with the two low-lying filaments
identified before the flare.
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE 9

4 (a) NLFFF (Top View) 4 (d) NLFFF (Top View)

2000

2 1000 2
0 (G)
Y (deg) 0 0

Y (deg)
−1000
−2 P1 −2000 −2 R1
P2 R3
N1 R2
−4 −4

−6 −6
Bottom: SHARP BZ 11:58:22 UT Bottom: AIA 1600 Å 12:47:53 UT (CEA)
−5 0 5 10 −5 0 5 10
X (deg) X (deg)
1 (b) NLFFF (Top View) 1 (e) NLFFF (Top View)
5
0 0 4

−1 −1 3
Y (deg)

Y (deg)
2
−2 −2
1
−3 −3
Outer Fan−Spine
−4 Inner Fan−Spine −4
NS Flux Rope
−5 EW Flux Rope −5 Bottom: logQ (z=0)
0 2 4 6 8 0 2 4 6 8
X (deg) X (deg)
1 2 3 4 5
(c) NLFFF (Side View)
(f) logQ (Vertical)
25
Spine
20
Z (Mm)

15
10 Fan
5
0
0 10 20 30 40 50 60
Distance (Mm)

Figure 5. NLFFF extrapolated magnetic field lines showing fan-spine and flux rope structures, which are overlaid on the
boundary SHARP Bz magnetogram (a)–(c), an AIA 1600 Å CEA map (d), and a filtered map of the squashing factor Q on
the bottom boundary (e), respectively. Also shown is a vertical Q-map revealing a fan-spine QSL structure (f). The field lines
are color-coded according to the legends in panel (b), and the Q-map is shown for the vertical cut along the orange dashed line
plotted in panel (e). In the left column, panels (b) and (c) show zoomed-in top and side views of the field lines, respectively.
Note that the coordinate origin in the top-view panels corresponds to the projection center of the original SHARP CEA map,
which has a Carrington coordinate of (220.28◦ , 17.39◦ ).
In Figure 5(d) we overlay the above magnetic field the chromospheric flare ribbons. The footpoints of the
lines on an AIA 1600 Å image taken at the second peak inner and outer spines are located exactly in the com-
in this passband (12:47:53 UT) when all the flare ribbons pact ribbon R3 and the eastern remote ribbon R1, re-
are clearly seen. For a better comparison between them, spectively. The degree of match found here seems bet-
we remap the AIA 1600 Å image into the same CEA ter than those in previously reported cases where the
projection as that of the SHARP map. Note that such spatial discrepancy between the outer spine footpoint
remapping will introduce artificial distortions to above- and the remote flare ribbon is typically over 10 Mm
the-surface structures, especially for those located far (e.g., Vemareddy & Wiegelmann 2014; Yang et al. 2015;
away from the disk center, just like the late-phase loops Masson et al. 2017). Except for the northernmost sec-
observed in this flare. For this reason, we do not at- tion, the circular ribbon R2 also matches the base of the
tempt to compare the extrapolated magnetic field with fan dome very well. In addition, the two footpoints of
the flare loops seen in other AIA coronal passbands. As the NS flux rope are anchored at ribbons R2 and R3,
revealed in Figure 5(d), there is a good match between respectively.
the footpoints of the overlaid magnetic field lines and
10 Dai et al.
The null point and the fan-spine structure indicate
the presence of magnetic separatrix and QSLs, whose
distribution can be quantified by the so-called squash-
ing factor Q (Titov et al. 2002). The Q factor quantita-
tively characterizes the connectivity gradients between
one magnetic field line and its neighboring field lines.
Here we use the method proposed by Pariat & Démoulin
(2012) to compute the distribution of Q value in the 3D
domain. In Figure 5(e) we plot the distribution of log Q
on the bottom boundary (z = 0). For the sake of clar-
ity, we apply a filter to the Q-map so that only the
distribution of log Q in regions of |Bz | >100 G is dis-
played. When overlaying the modeled field lines on the
Q-map, it is clearly seen that the fan-spine field lines
intersect with the solar surface at patches of high log Q
values. This is consistent with the observational fact
found in many previous studies (e.g., Chen et al. 2012;
Masson et al. 2017) that flare ribbons typically stop at
the intersection of QSLs with the solar surface. To de-
pict the structure of the QSLs in the corona, we draw a
slice across the fan-dome base and the footpoint of the
inner spine on the bottom boundary (the orange dashed
line in Figure 5(e)), and plot the 2D Q-map in the ver-
tical plane extending above the slice, which is shown in
Figure 5(f). As expected, it clearly shows the fan-spine
magnetic topology of this event, i.e., a QSL surrounding
the fan-spine separatrix (Masson et al. 2009).
There are still some discrepancies in morphology be-
tween the extrapolated magnetic field and the flare rib-
bons. First, the northernmost section of the circular rib-
bon R2 extends well beyond the fan-dome base, showing
a spatial offset up to 15 Mm (Figure 5(d)). Second, in
spite of the excellent match between the outer spine foot-
point and the eastern remote ribbon R1, field lines that
originate from other parts of this elongated ribbon, pass
through the fan-spine QSL, and end up at the circular
ribbon R2 cannot be reasonably figured out, although
their existence can be confirmed by the appearance of
the widespread late-phase loops. For this reason, we do
not display these field lines in Figure 5. As mentioned
above, at the time of flare, the AR is located far away
from the disk center. The SHARP remapping procedure
could hence amplify the errors of magnetic field mea-
sured in the local plane, which are then transferred to
all the field components at the bottom boundary used
for extrapolation. This may significantly compromise
the fidelity of magnetic field extrapolation from regions
of relatively weak field strengths.
4. MODELING OF THE LATE-PHASE LOOPS
4.1. Stereoscopic Measurement of the Loop Lengths
By combing the full-disk integrated EVE and spatially
resolved AIA observations, a clear correspondence be-
tween the late-phase emissions and the long late-phase
loops in this flare has been validated. The length of
a coronal loop is, both theoretically and observation-
ally, an important parameter for the loop evolution. In
previous case studies, estimation of the lengths of late-
phase loops was mainly based on single-perspective ob-
servations and therefore subject to large uncertainties.
Through imaging observations from multiple perspec-
tives, we could in principle track the realistic coronal
loops in 3D, and hence more accurately determine the
loop lengths.
Figure 6(a) depicts the deployment of the Solar Ter-
restrial Relations Observatory (STEREO ; Kaiser et al.
2008) twin spacecraft with respect to the Earth at the
time of flare. The longitudinal position of the flare site
(indicated by the black dot) and its favorable viewing
angles to STEREO-A and near-Earth SDO enable us
to stereoscopically measure the 3D coordinates of flare
loops captured by both satellites. In Figures 6(b) and
(c) we display paired images of the late-phase loops
taken simultaneously at ∼13:50 UT by the EUV Im-
ager (EUVI; Wuelser et al. 2004) on board STEREO-A
(hereafter EUVI-A) in 195 Å and by AIA in 193 Å,
respectively, both of which have a similar temperature
response peaking around 1.3 MK. As shown in the fig-
ures, the late-phase loops appear mainly in two clusters.
In the north, the late-phase loops are readily resolved
in both passbands. Using the Solar Software (SSW;
Freeland & Handy 1998) routine scc measure.pro, we
trace a full late-phase loop, whose 3D coordinates are
projected onto the two perspectives (dashed lines), re-
spectively. By integrating the distance along the loop,
we derive a length of 81 Mm for this late-phase loop.
In the south, the triangulation measurement of the late-
phase loops is nevertheless significantly affected by an
LOS effect. When viewed from the two different per-
spectives, either the eastern (from EUVI-A) or the west-
ern (from AIA) legs of the late-phase loops coincidently
overlap along the corresponding LOS. As a result, only
the legs on the opposite side are distinguishable from the
corresponding perspective, as pointed out by the arrows
in the figures. Moreover, the intensity of the loop-top
regions in both passbands is predominately high, since
the intensity observed there is actually an integration
along the LOS, which is also contributed by emissions
from the loop legs. Restricted by the LOS effect, here
we just stereoscopically estimate the apex location of
the late-phase loops (indicated by the asterisks), whose
height is ∼35 Mm above the solar surface. Assuming a
semi-circular loop geometry, we obtain a loop length of
∼110 Mm. Note that due to the uncertainty in deter-
mining the loop apex, this value might be an underesti-
mation of the lengths for the southern late-phase loops.
The projections of the axis of the late-phase loop in
the north and apex of the late-phase loop in the south
are further overplotted in an AIA 335 Å difference image
(Figure 6(d)) taken 13 minutes earlier (at the late-phase
peak in AIA 335 Å). We find an excellent match between
the loop geometries and emission features in the two AIA
passbands, further confirming the slow cooling process
in the long late-phase loops.
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE 11

(a) (b) EUVI−A 195 Å

−1.0

250
−0.5
A
HEE_Y (AU)

Y (arcsec)
B Sun 200
0.0
Flare

0.5 150
Western Legs

1.0
Earth 100 13:50:30 UT

−1.0 −0.5 0.0 0.5 1.0 −850 −800 −750 −700

HEE_X (AU) X (arcsec)

(c) AIA 193 Å (d) AIA 335 Å

250 250

200 200
Y (arcsec)

Y (arcsec)

150 150
1 2 1 2

Eastern Legs Eastern Legs

100 100
13:50:43 UT 13:37:39 UT

650 700 750 800 650 700 750 800

X (arcsec) X (arcsec)

Figure 6. Deployment of the STEREO twin spacecraft with respect to the Earth at the time of flare (a), and images of the
late-phase loops viewed from two different perspectives (b)–(d). In panel (a), the black dot indicates the longitudinal position of
the flare site. In panels (b)–(d), the dashed lines and asterisks outline the projection of the characteristic late-phase loops onto
the two different perspectives, respectively, and the arrows point to the legs of the southern late-phase loops. In the bottom
row images, two boxes are drawn at the leg and footpoint of a southern late-phase loop, respectively, over which we derive the
AIA light curves used for modeling of the late-phase emissions.
For comparison, we also quantify the lengths of the 4.2.1. Model Setup
main flaring loops, which are simply inferred from the
separation distance between the flare ribbons R2m and The EBTEL model is a zero-dimensional (0D) hydro-
R3 on which the two conjugate loop footpoints are an- dynamic model that describes the evolution of the aver-
chored. The estimated lengths of the main flaring loops age temperature, pressure, and density along a coronal
are 10–25 Mm. Obviously, the two sets of flare loops loop (or more strictly speaking, a coronal strand). The
are distinct in length, with a length ratio between the basic idea behind EBTEL is that the imbalance between
late-phase loops and the main flaring loops exceeding 3. the heat flux conducting down into the TR and the ra-
diative loss rate there leads to an enthalpy flux at the
coronal base: an excess heat flux drives an evaporative
4.2. EBTEL Modeling of the Late-Phase Emissions upflow, whereas a deficient heat flux is compensated for
The electromagnetic emissions from a flaring loop re- by a condensation downflow. This upward/downward
flect the underlying evolution of temperature and den- enthalpy flow controls the hydrodynamic evolution of
sity, and thereby the energy release history in the loop. the coronal loop.
Using the EBTEL model, we model the EUV emissions Another assumption of EBTEL is that any flows along
of a late-phase loop in this flare. By comparing the syn- the loop are subsonic, which should hold for most of
thetic EUV light curves with the observed light curves the time of the loop evolution. This means that the
in several AIA passbands, we quantify parameters of the kinetic energy is negligible compared to the internal en-
heating on the late-phase loop as well as the loop geom- ergy. Hence the equation of momentum can be dropped
etry. from the governing equations, so do the kinetic terms
12 Dai et al.

Norm. AIA Intensity

1.0 (a) 1600 Å
from the equation of energy conservation. Following this Observed
Gaussian Fit
0.8 Box 1
way, in EBTEL we solve the remaining loop-averaged 0.6
t0=12:43:08 UT
τ=23 s
(along the coronal section of the loop) hydrodynamic 0.4
equations including the equations of continuity and en- 0.2
0.0
ergy conservation, and equation of state, which are given
12:40 12:44 12:48 12:52 12:56 13:00
by 1000
dn c2 (b) 131 Å
800 Box 2
Observed
=− (F0 + c1 Rc ), (1) Synthetic
dt 5c3 kT L 600
  400
dp 2 Rc 200
= H − (1 + c1 ) , (2)
dt 3 L 0
250
(c) 94 Å
and 200 Box 2
1 dT 1 dp 1 dn 150

AIA Intensity (DN s−1 pix−1)

= − , (3)
T dt p dt n dt 100
50
where n, p, and T are the average density, pressure, 0
and temperature of the coronal loop, respectively, L is 120 (d) 335 Å H0=1.1 erg cm−3 s−1
100 Box 2 L=82 Mm
the loop half-length (EBTEL assumes a symmetric semi- 80
WL=5.2 Mm
circular loop geometry and hence treats only half of the 60
40
loop), k is the Boltzmann constant, c2 (c3 ) is the ra- 20
tio of the average coronal (coronal base) temperature to 0
apex temperature, F0 = −(2/7)κ0 (T /c2 )7/2 /L (where 3000 (e) 171 Å
Box 2
κ0 = 8.12×10−7 in cgs units is the classical Spitzer ther- 2000
mal conduction coefficient) is the heat flux at the coronal
1000
base, Rc = n2 Λ(T )L (where Λ(T ) is the optically thin
radiative loss function) approximates the radiative loss 0

rate from the corona, c1 = Rtr /Rc is the ratio of radia-
tive loss rate of the TR to that of the corona, and H
denotes the average volumetric heating rate (note that
here we use H instead of Q, the commonly adopted nota-
tion for heating rate in the literature, to avoid confusion
with the squashing factor introduced in Section 3).
Of the three dimensionless parameters c1 , c2 , and c3
in EBTEL, c1 is the most important. The main dif-
ference between the different versions of EBTEL lies in
the treatment of c1 . Assuming a single power-law radia-
tive loss function, Martens (2010) analytically studied
the energy equilibrium for a static coronal loop, whose
solutions reveal a value of c1 being around 2, and val-
ues of c2 and c3 close to 0.9 and 0.6, respectively. In
the original EBTEL model (Klimchuk et al. 2008), to
achieve an overall consistency with the 1D simulation
results, c1 was kept as a constant of 4 throughout the
loop evolution. Nevertheless, when a loop evolves dy-
namically, the equilibrium assumption is broken and c1
also evolves. Obviously, the choice of a fixed c1 value is
not physically reasonable. In the later EBTEL versions,
more physics was included to appropriately determine
the values of c1 . One piece of physics previously ig-
nored is gravitational stratification, whose main effect
is to depress the coronal radiation. Thus, higher val-
ues of c1 can be expected, especially for long loops with
large ratios of the length to the gravitational scale height
(Cargill et al. 2012). Another piece is the deviation of
the dynamically cooling loop from equilibrium states.
In the early conductive cooling phase, the density in-
crease in response to the strong heat flux is not so fast
that the loop is under-dense with respect to a static loop
12:40 13:00 13:20 13:40 14:00 14:20 14:40
Time (UT on 2011−Sep−09)
Figure 7. Top: Normalized time profile observed at the
footpoint of the selected late-phase loop (Box 1 in Figure 6)
in the AIA 1600 Å passband (solid curve), with the Gaus-
sian fit overplotted (dashed curve). Bottom: Light curves
observed at the leg of the same loop (Box 2 in Figure 6)
in four AIA coronal passbands (solid curves), as well as the
synthetic light curves (dashed curves) computed with the
best-fit parameters in the EBTEL modeling (specified in the
legends in panel (d)).
at the same temperature, leading to higher values of c1
(Barnes et al. 2016). During the radiative cooling phase,
the loop is instead over-dense, which in turn reduces the
c1 values (Cargill et al. 2012). In our modeling, we hold
c2 and c3 at typical values of 0.9 and 0.6, respectively, as
specified in all EBTEL versions, whereas we use the lat-
est EBTEL version (Barnes et al. 2016) to consistently
calculate the c1 values.
4.2.2. Comparison with Observations
Based on the output temperature and density from
the EBTEL model, we can synthesize the emissions from
the loop. The synthetic intensity of the loop in an AIA
passband can be computed as
IAIA = R(T )n2 WL (DN s−1 pix−1 ), (4)
where R(T ) is the temperature response function of the
AIA passband, which can be obtained by the SSW rou-
tine aia get response.pro, and WL is the nominal
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE
width (depth) of the loop along the LOS. In gener-
ating the AIA response functions, we set the evenorm
keyword, which will give better agreement with full-disk
EVE observations when cross-checking data from both
AIA and EVE. Due to the 0D nature of the EBTEL
model, it is physically not applicable to give a spatial
distribution of the synthetic intensities along the loop.
To compare the synthetic intensities with observations,
we select a late-phase loop in the south whose observed
intensities are roughly uniformly distributed along its
eastern leg, as shown in the bottom row AIA images
of Figure 6. We draw a 2′′ × 2′′ box over the loop leg
(Box 2 in Figure 6). The intensities averaged over this
box in several AIA coronal passbands are taken as the
observed loop intensities, whose time profiles are shown
as the solid curves in Figures 7(b)–(e).
To infer the heating function for the loop, we calcu-
late the chromospheric light curve at the loop footpoint,
as proposed by Li et al. (2012), Qiu et al. (2012), and
Zhu et al. (2018). To this end, we draw a 6′′ × 6′′ box
(Box 1 in Figure 6) at the eastern footpoint of the se-
lected late-phase loop, which is located on the remote
flare ribbon R1. In Figure 7(a) we plot the normalized
AIA 1600 Å intensity profile averaged over this box (the
solid curve), which closely resembles the intensity profile
for the whole eastern ribbon shown in Figure 3(e). By
applying a Gaussian fit to the rise and fast decay phase
of the AIA 1600 Å light curve, we derive a peak time at
12:43:08 UT and a width of 23 s for the fitted Gaussian
function (the dashed curve in Figure 7(a)). Mimicking
the shape of the fitted Gaussian function, the imposed
impulsive heating on the late-phase loop is then given
by
(t − t0 )2
 
H(t) = H0 exp − (erg cm−3 s−1 ), (5)
2τ 2
where H0 is the amplitude of the heating, t0 =12:43:08 UT
the time of the peak heating rate, and τ =23 s the Gaus-
sian duration of the heating. Note that there is a second
small peak ∼5 minutes after the first peak in the AIA
1600 Å intensity profile, which possibly reflects a sec-
ond heating process. Nevertheless, this second heating
should be much weaker than the first heating, and hence
can be safely ignored in our modeling. In addition to the
impulsive heating, we also include a steady background
heating whose rate is fixed to be 10−5 erg cm−3 s−1 .
While maintaining the loop temperature and density
before and long after the impulsive heating, this back-
ground heating does not obviously affect the modeled
EUV emissions from the loop.
There are three free parameters in our modeling, i.e.,
H0 , L, and WL . Although the characteristic loop length
for the southern late-phase loops has been estimated us-
ing the triangulation method, the LOS effect may intro-
duce some uncertainties. Therefore, here we still regard
L as a free parameter. We use the MPFIT (Markwardt
2009) package distributed in SSW to perform the fitting.
13
In our modeling, we only fit the observed light curve of
the late-phase loop in AIA 335 Å, since we mainly focus
on the EUV late phase in the warm coronal emissions.
For other passbands, we manually inspect the degree of
match between the synthetic light curves computed with
the best-fit parameters and the observed light curves.
Some factors that could affect the loop emissions in these
passbands are addressed later in this section.
The fitting gives best-fit values for the three free
parameters: H0 =1.1 erg cm−3 s−1 , L=82 Mm, and
WL =5.2 Mm. To test the robustness of the fitting, we
vary the starting guesses for the parameters within a
reasonable range. In most cases, the fitting procedure
converges to the same results as those given above. The
synthetic light curves computed with the best-fit pa-
rameters are overplotted as the dashed curves in Fig-
ures 7(b)–(e). As seen in Figure 7(d), the peak in both
the synthetic and the observed light curves in AIA 335 Å
occurs almost one hour after the impulsive heating. In
the other passbands, the peak times in the synthetic
light curves also reasonably agree with those in the ob-
served light curves. This indicates that even without
an additional delayed heating, the long cooling process
of the late-phase loop can sufficiently explain the pres-
ence of the late-phase emissions at different tempera-
tures. Note that there are three consecutive peaks in
the observed AIA 171 Å light curve. It means that the
LOS should intersect at least three late-phase loops of
similar half-lengths, and the parameter WL in the mod-
eling would reflect a combination of the widths of all
these loops seen along the LOS. In this sense, the ef-
fective width of an individual late-phase loop is of the
order of 1.7 Mm, which is consistent with those found
in previous observations (e.g., Aschwanden et al. 2008).
In spite of a good match between the modeled results
and observations seen in AIA 335 Å, there is a large
discrepancy in amplitude between the synthetic light
curves and the observed ones in the other passbands. In
the hotter AIA 131 and 94 Å passbands, the synthetic
peak intensities are significantly higher than those ob-
served. The hotter the passband, the more prominent
such a discrepancy (Figures 7(b) and (c)). We attribute
this discrepancy to a possible deviation from an ioniza-
tion equilibrium in the late-phase loops, a factor not
considered in our EBTEL modeling. During the im-
pulsive heating and conductive cooling stage, the loop
temperature evolves so fast that the ionization process
cannot catch up with the temperature change, result-
ing in a non-equilibrium ionization (Reale & Orlando
2008; Bradshaw & Klimchuk 2011). Obviously, this ef-
fect is the most prominent for hot coronal emissions from
long tenuous loops, e.g., the late-phase loops. Using
the full 1D HYDRAD code (Bradshaw & Mason 2003),
Bradshaw & Klimchuk (2011) numerically studied the
effect of non-equilibrium ionization on loop emissions,
and found that the synthetic loop intensities in hot pass-
bands such as AIA 131 Å can be depressed by orders of
14 Dai et al.
magnitude compared to those under the ionization equi-
librium assumption. They also pointed out that as the
cooling rate slows down and the loop density increases,
the ions can have enough time to reach an ionization
equilibrium at a medium temperature like 6 MK. There-
fore, emissions formed below this temperature (like the
EUV late-phase emission in AIA 335 Å) are unlikely to
be notably affected by this effect. Another piece of evi-
dence for the existence of non-equilibrium ionization in
the modeled late-phase loop is the small bump seen in
the synthetic AIA 335 Å light curve shortly after the im-
pulsive heating, which is also synchronous with the peak
of the synthetic light curve in AIA 131 Å. This bump
reflects a high temperature response over 10 MK in AIA
335 Å, which is introduced by crosstalk from the AIA
131 Å passband (Boerner et al. 2012). Nevertheless, due
to the possible non-equilibrium ionization at flare tem-
peratures early in the long tenuous late-phase loop, the
bump is absent in the observed AIA 335 Å light curve.
Recently, Zhu et al. (2018) studied a C-class two-ribbon
solar flare using the EBTEL model, in which the mod-
eled peaks in AIA 131 and 94 Å are also well above the
observed peaks (Figure 3 in their paper).
On the other hand, the synthetic loop intensities in
the cooler AIA 171 Å passband are systematically lower
than the observed intensities (Figure 7(e)). It implies
an over-estimation of the mass draining from the corona
during the radiative cooling phase. The reason may lie
in the geometry of the late-phase loop. We note that the
best-fit value of parameter L for the southern late-phase
loop is 82 Mm in the EBTEL modeling, which is signif-
icantly higher than the value of ∼55 Mm inferred from
the height of the loop apex based on the semi-circular
loop assumption. Such a large discrepancy in length is
unlikely to be simply ascribed to the uncertainty in de-
termining the loop height, while it is more likely due to
an oblate shape of the late-phase loop rather than the
semi-circular geometry as commonly assumed. With the
same loop length, a semi-circular loop can reach a higher
altitude than an oblate loop. Therefore, in a semi-
circular loop, the coronal emission is more significantly
depressed by the gravitational stratification, which ac-
cordingly leads to a higher c1 value. Since the EBTEL
model assumes a semi-circular loop geometry for sim-
plicity, it will overestimate the c1 values when dealing
with an actually oblate-shaped loop, and hence inap-
propriately elevate the mass draining rate during the
radiative cooling phase in order to compensate for the
over-estimated TR radiative loss (Cargill et al. 2012).
As a result, the modeled loop intensities during this
stage will be under-estimated, as is seen in AIA 171 Å
for the late-phase loop modeled here. At higher tem-
peratures, for example, those corresponding to the AIA
335 Å passband, the relative mass draining rate (quan-
tified as d ln n/d ln T ) is not so fast that the relevant
loop emission would be much less impacted. In another
EBTEL modeling of an M-class solar flare by Li et al.
(2012), the modeled loop intensities in AIA 211 and
171 Å are also systematically lower than the observed in-
tensities (Figure 4 in their paper). We note that Li et al.
(2012) fixed the parameter c1 for each individual loop,
whose value would be too high for the radiative cooling
phase.
5. DISCUSSION
Based on the observations and the NLFFF extrapo-
lation results, we propose a two-stage magnetic recon-
nection scenario to explain the observed flare evolution.
The flare is preceded by a brightening first occurring
around or within the EW flux rope (manifested by the
EW filament), which then propagates eastward and trig-
gers an instability of the nearby NS flux rope (mani-
fested by the NS filament). The details of the trigger-
ing process are beyond the scope of this work. Here
we conjecture that the helical kink instability may be
responsible for the fast ascent of the initially low-lying
NS flux rope. As the flux rope rises up, the first-stage
magnetic reconnection sets in. Inside the flux rope,
possible reconnection due to an internal kink instabil-
ity (Galsgaard & Nordlund 1997) can heat the filament
body, and more importantly, accelerate high-energy par-
ticles that produce the two conjugate HXR sources at
the footpoints of the flux rope, as proposed in Guo et al.
(2012). Meanwhile, the flux rope pushes the overlying
field lines toward the null point, and makes the fan-
spine QSL structure thinner, causing an enhancement
the current inside it. Reconnections both around the
null point and in the QSL can produce the observed mul-
tiple flare ribbons and long late-phase loops connecting
the eastern remote ribbon R1 and circular ribbon R2.
Considering the elongation of the remote ribbon R1 and
the appearance of the late-phase loops, the QSL recon-
nection seems more energetic than the null-point recon-
nection. Moreover, since all flare ribbons brighten up
very impulsively, we suggest that the QSL reconnection
could belong to a super-Alfvénic slip-running reconnec-
tion (Aulanier et al. 2006), during which the inferred
elongation motion of ribbon brightening has been com-
plete within the time cadence of AIA.
As the fast ascending flux rope slows down, the
second-stage magnetic reconnection starts to be ini-
tiated. Reconnection mainly takes place in the current
sheet formed between the two legs of the overlying field
lines stretched by the flux rope. The newly formed rib-
bon segment R2c fills the gap in the circular ribbon R2
formed in the first stage. Short flaring loops connect
R2c with the compact ribbon R3. Such a reconnection
is predicted by the standard 2D CSHKP flare model,
and the observed separation speed of the flare ribbon
(∼30 km s−1 after correction for the projection effect)
is also typical for a two-ribbon flare. Therefore, we refer
to the observed flaring loops as “main flaring loops”.
Meanwhile, there also exists a less energetic QSL recon-
nection, which is responsible for the elongation motion
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE
of small brightening along the remote ribbon R1. The
revealed motion speed of 180 km s−1 implies that the
QSL reconnection during this stage may correspond
to a sub-Alfvénic slipping reconnection (Aulanier et al.
2006).
Based on the two-stage reconnection scenario, the long
late-phase loops are mainly heated during the first-stage
QSL reconnection, while the second-stage flare reconnec-
tion, occurring ∼5 minutes later, is responsible for the
heating of short main flaring loops. This process cannot
be accounted for by the standard CSHKP flare model, in
which shorter flare loops are always heated earlier than
longer loops. Nevertheless, it is a natural consequence
of a multipolar magnetic topology, as is the case in this
flare. We note that our scenario is basically in agreement
with the picture proposed by Sun et al. (2013) based on
a toy model (Figure 1 in their paper) except that in their
opinion the late-phase emission mainly comes from the
outer spine lines and is caused by reconnection at the
fan-spine null point.
The close correlation between the rise of the flux rope
and the brightening of the flare ribbons indicates that
the flux rope should be the main driving agent for the
reconnections in both stages. During the first stage, the
fast ascending flux rope makes the above QSL recon-
nection more enhanced than usual. Nevertheless, this
QSL reconnection does not efficiently remove the strong
magnetic constraints above the flux rope, which there-
fore slows down and finally fails to erupt. The field lines
involved in the second-stage reconnection have not yet
been sufficiently stretched when the flux rope stops ris-
ing, causing a significant depression of the reconnection
rate in this stage. Therefore, it is likely that more en-
ergy is injected to the late-phase loops than the main
flaring loops, leading to an extremely large EUV late
phase in this non-eruptive flare.
We also quantify the heating history in a typical late-
phase loop. The chromospheric AIA 1600 Å light curve
observed at the loop footpoint further confirms that the
late-phase loop is heated by an impulsive energy re-
lease during the first-stage reconnection. By using the
EBTEL model, we synthesize the loop intensities and
compare them with observations. Fitting the observed
intensities with the synthetic ones yields best-fit values
for the peak heating rate H0 =1.1 erg cm−3 s−1 and
loop half-length L =82 Mm. Due to a contamination
from the TR and chromosphere, we cannot reliably ex-
tract the loop intensities for the much shorter main flar-
ing loops. Furthermore, we cannot reliably characterize
the contribution of nonthermal electron beam heating,
which is believed to be marginal in the heating of late-
phase loops but becomes important during main flare
heating. These factors prevent us from applying the
same EBTEL modeling to the main flaring loops and
making a direct comparison of the heating rate between
the two sets of loops. Nevertheless, when comparing our
results with those revealed in other flares of the similar
15
GOES classes (e.g., Li et al. 2012; Zhu et al. 2018), it
is found that the peak heating rate of the late-phase
loops in this flare is comparable to or even greater than
the heating rate of the main flaring loops in those flares.
Considering the larger length of the late-phase loop, it is
suggested that the late-phase loop undergoes an intense
heating with a peak energy deposition rate (2H0 L) as
high as 1.8×1010 erg cm−2 s−1 and total energy input
of 7.1×1028 erg.
For a late-phase loop with a half-length of 82 Mm,
the loop emission in AIA 335 Å peaks almost one hour
after the impulsive heating. Since there is no obvious
additional heating on the late-phase loop, the time de-
lay of the warm coronal emission peak is mainly due
to the long cooling process of the late-phase loop, as
proposed in Liu et al. (2013). By comparison, the main
flaring loops are significantly shorter (2L∼10–25 Mm),
and hence evolve more quickly. In a numerical experi-
ment using the EBTEL model, Dai & Ding (2018) found
that with a strong initial loop heating, the hot and warm
coronal emissions peak in the conductive cooling and ra-
diative cooling stages, respectively. For the hot coronal
emissions, the main flare peak and the late-phase peak
temporally overlap because the cooling rate in all loops
is very fast during the conductive cooling phase. How-
ever, the small difference in cooling rate between the
loops of different lengths can still produce an observable
bump following the main peak, which results in a dual-
decay behavior in the corresponding light curves, as seen
in this flare (Figures 1(a) and (c)). We note that this
dual-decay in GOES SXRs has been used as a proxy
for EUV late-phase flares prior to the SDO era (Woods
2014). When the loops continue to cool down, the effect
of the difference in loop length on the loop cooling time
becomes more and more prominent. At warm coronal
temperatures, the late-phase peak can be well separated
and thus easily distinguished from the main flare peak.
Since the warm coronal emission peaks in the radiative
cooling stage, the emission after the peak can be notably
affected by the mass draining, hence showing a fast de-
cay compared to the relatively gradual rise. This pattern
is consistent with the shape of the light curves of AIA
335 Å in this flare, both for the whole AR (Figure 1(e))
and for the individual late-phase loop (Figure 7(d)).
In the EBTEL modeling of the late-phase loop, the
synthetic peak intensities in the hotter AIA passbands
are significantly higher than those observed. We at-
tribute this discrepancy to a non-equilibrium ionization
effect (Reale & Orlando 2008; Bradshaw & Klimchuk
2011) in the late-phase loop. To further verify this hy-
pothesis, we perform a new set of fitting in which the
observational data from both AIA 335 and 94 Å are
fitted, as done in Li et al. (2012). In this approach,
no matter how we change the starting guesses for the
parameters, the synthetic peak in AIA 94 Å based on
the “best-fit” parameters is still systematically above
the observed peak, although the discrepancy between
16 Dai et al.
them is somewhat reconciled, and more importantly,
the synthetic peak in AIA 335 Å is now systematically
lower than the observed peak (not shown here). Since
the light curve in neither AIA 335 Å nor AIA 94 Å can
be reasonably modeled, the even worse fitting results
indicate that the late-phase emissions observed in AIA
94 Å (and also in AIA 131 Å) are indeed affected by the
effect of non-equilibrium ionization. This also explains
why an EUV late phase as large as that in AIA 335 Å is
not observed in a hotter passband, e.g., AIA 94 Å, even
though the late-phase peak is clearly separated from the
main flare peak (Figure 1(d)).
In the cooler AIA passbands, the higher mass draining
rate during the radiative cooling stage can also cause the
late-phase peak less prominent (see Figure 1(f)). On the
other hand, we find that the synthetic intensities of the
late-phase loop in AIA 171 Å are systematically lower
than the observed intensities. We propose that the dis-
crepancy may be caused by a semi-circular assumption
of the loop geometry in EBTEL, which is not appropri-
ate for the real geometry of the late-phase loop modeled
here.
6. SUMMARY
We have analyzed and modeled an M1.2 non-eruptive
solar flare on 2011 September 9 that exhibits an ex-
tremely large EUV late phase in the warm coronal emis-
sions. Based on an NLFFF extrapolation, we proposed a
two-stage magnetic reconnection scenario to explain the
REFERENCES
Antiochos, S. K., & Sturrock, P. A. 1976, SoPh, 49, 359
—. 1978, ApJ, 220, 1137
Asai, A., Yokoyama, T., Shimojo, M., et al. 2004, ApJ, 611,
557
Aschwanden, M. J., Nitta, N. V., Wuelser, J.-P., & Lemen,
J. R. 2008, ApJ, 680, 1477
Aulanier, G., Pariat, E., Démoulin, P., & DeVore, C. R.
2006, SoPh, 238, 347
Aulanier, G., Golub, L., DeLuca, E. E., et al. 2007, Science,
318, 1588
Barnes, W. T., Cargill, P. J., & Bradshaw, S. J. 2016, ApJ,
829, 31
Bobra, M. G., Sun, X., Hoeksema, J. T., et al. 2014, SoPh,
289, 3549
Boerner, P., Edwards, C., Lemen, J., et al. 2012, SoPh, 275,
41
Bradshaw, S. J., & Klimchuk, J. A. 2011, ApJS, 194, 26
Bradshaw, S. J., & Mason, H. E. 2003, A&A, 401, 699
Brueckner, G. E., Howard, R. A., Koomen, M. J., et al.
1995, SoPh, 162, 357
Calabretta, M. R., & Greisen, E. W. 2002, A&A, 395, 1077
evolution of the flare, in which different reconnections
are involved and take place in different places. Using
the EBTEL model, we modeled the EUV emissions from
a late-phase loop. The modeling reveals a high heating
rate for the late-phase loop. Our main conclusion is that
the extremely large late phase is powered by an intense
heating even earlier than the main flare heating, and the
delayed occurrence of the late-phase peak is mainly due
to the long cooling process of the long late-phase loops.
The production of EUV late phase in this flare is dif-
ferent from that in the 2011 September 6 X2.1 eruptive
flare not only in timing of the heating process but also
in the heating rate. In addition to these two flares, AR
NOAA 11283 has produced a series of EUV late-phase
flares, which exhibit different evolution patterns in the
late-phase emissions. A comparative study of the pro-
duction of the late phase in these flares will be presented
in a following work.
We are very grateful of the anonymous referee whose
valuable comments and suggestions led to a significant
improvement of the manuscript. This work was sup-
ported by National Natural Science Foundation of China
under grants 11533005 and 11733003, and 973 Project
of China under grant 2014CB744203. D.Y. is also spon-
sored by the Open Research Project of National Center
for Space Weather, China Meteorological Administra-
tion. SDO is a mission of NASA’s Living With a Star
(LWS) Program.
Cargill, P. J., Bradshaw, S. J., & Klimchuk, J. A. 2012,
ApJ, 752, 161
Cargill, P. J., Mariska, J. T., & Antiochos, S. K. 1995, ApJ,
439, 1034
Carmichael, H. 1964, NASA Special Publication, 50, 451
Chamberlin, P. C., Milligan, R. O., & Woods, T. N. 2012,
SoPh, 279, 23
Chen, P. F., Su, J. T., Guo, Y., & Deng, Y. Y. 2012,
Chinese Science Bulletin, 57, 1393
Dai, Y., & Ding, M. 2018, ApJ, 857, 99
Dai, Y., Ding, M. D., & Guo, Y. 2013, ApJL, 773, L21
Démoulin, P., Hénoux, J. C., Priest, E. R., & Mandrini,
C. H. 1996, A&A, 308, 643
Feng, L., Wiegelmann, T., Su, Y., et al. 2013, ApJ, 765, 37
Freeland, S. L., & Handy, B. N. 1998, SoPh, 182, 497
Galsgaard, K., & Nordlund, Å. 1997, J. Geophys. Res., 102,
219
Gary, G. A., & Hagyard, M. J. 1990, SoPh, 126, 21
Guo, Y., Ding, M. D., Schmieder, B., Démoulin, P., & Li,
H. 2012, ApJ, 746, 17
Hirayama, T. 1974, SoPh, 34, 323
 EUV LATE PHASE IN A NON-ERUPTIVE FLARE
Hock, R. A., Woods, T. N., Klimchuk, J. A., Eparvier,
F. G., & Jones, A. R. 2012, ArXiv e-prints,
arXiv:1202.4819
Hoeksema, J. T., Liu, Y., Hayashi, K., et al. 2014, SoPh,
289, 3483
Jiang, C., Wu, S. T., Feng, X., & Hu, Q. 2014, ApJ, 780, 55
Kaiser, M. L., Kucera, T. A., Davila, J. M., et al. 2008,
SSRv, 136, 5
Kane, S. R., & Donnelly, R. F. 1971, ApJ, 164, 151
Klimchuk, J. A., Patsourakos, S., & Cargill, P. J. 2008,
ApJ, 682, 1351
Kopp, R. A., & Pneuman, G. W. 1976, SoPh, 50, 85
Lau, Y.-T., & Finn, J. M. 1990, ApJ, 350, 672
Leka, K. D., Barnes, G., Crouch, A. D., et al. 2009, SoPh,
260, 83
Lemen, J. R., Title, A. M., Akin, D. J., et al. 2012, SoPh,
275, 17
Li, Y., Ding, M. D., Guo, Y., & Dai, Y. 2014, ApJ, 793, 85
Li, Y., Qiu, J., & Ding, M. D. 2012, ApJ, 758, 52
Lin, R. P., Dennis, B. R., Hurford, G. J., et al. 2002, SoPh,
210, 3
Liu, C., Deng, N., Lee, J., et al. 2014, ApJ, 795, 128
Liu, K., Wang, Y., Zhang, J., et al. 2015, ApJ, 802, 35
Liu, K., Zhang, J., Wang, Y., & Cheng, X. 2013, ApJ, 768,
150
Markwardt, C. B. 2009, in Astronomical Society of the
Pacific Conference Series, Vol. 411, Astronomical Data
Analysis Software and Systems XVIII, ed. D. A.
Bohlender, D. Durand, & P. Dowler, 251
Martens, P. C. H. 2010, ApJ, 714, 1290
Masson, S., Pariat, E., Aulanier, G., & Schrijver, C. J.
2009, ApJ, 700, 559
Masson, S., Pariat, É., Valori, G., et al. 2017, A&A, 604,
A76
Metcalf, T. R., Leka, K. D., Barnes, G., et al. 2006, SoPh,
237, 267
Miklenic, C. H., Veronig, A. M., Vršnak, B., & Hanslmeier,
A. 2007, A&A, 461, 697
Milligan, R. O., Chamberlin, P. C., Hudson, H. S., et al.
2012, ApJL, 748, L14
Neupert, W. M. 1968, ApJL, 153, L59
17
O’Dwyer, B., Del Zanna, G., Mason, H. E., Weber, M. A.,
& Tripathi, D. 2010, A&A, 521, A21
Pariat, E., & Démoulin, P. 2012, A&A, 541, A78
Parker, E. N. 1963, ApJS, 8, 177
Pesnell, W. D., Thompson, B. J., & Chamberlin, P. C.
2012, SoPh, 275, 3
Priest, E. R., & Longcope, D. W. 2017, SoPh, 292, 25
Qiu, J., Liu, W.-J., & Longcope, D. W. 2012, ApJ, 752, 124
Qiu, J., Longcope, D. W., Cassak, P. A., & Priest, E. R.
2017, ApJ, 838, 17
Reale, F., & Orlando, S. 2008, ApJ, 684, 715
Romano, P., Zuccarello, F., Guglielmino, S. L., et al. 2015,
A&A, 582, A55
Ruan, G., Chen, Y., & Wang, H. 2015, ApJ, 812, 120
Scherrer, P. H., Schou, J., Bush, R. I., et al. 2012, SoPh,
275, 207
Sturrock, P. A. 1966, Nature, 211, 695
Sun, X., Hoeksema, J. T., Liu, Y., et al. 2013, ApJ, 778, 139
Titov, V. S., Hornig, G., & Démoulin, P. 2002, Journal of
Geophysical Research (Space Physics), 107, 1164
Vemareddy, P., & Wiegelmann, T. 2014, ApJ, 792, 40
Wang, Y., Zhou, Z., Zhang, J., et al. 2016, ApJS, 223, 4
Wheatland, M. S., Sturrock, P. A., & Roumeliotis, G. 2000,
ApJ, 540, 1150
Wiegelmann, T. 2004, SoPh, 219, 87
Wiegelmann, T., Inhester, B., & Sakurai, T. 2006, SoPh,
233, 215
Woods, T. N. 2014, SoPh, 289, 3391
Woods, T. N., Hock, R., Eparvier, F., et al. 2011, ApJ, 739,
59
Woods, T. N., Eparvier, F. G., Hock, R., et al. 2012, SoPh,
275, 115
Wuelser, J.-P., Lemen, J. R., Tarbell, T. D., et al. 2004, in
Proc. SPIE, Vol. 5171, Telescopes and Instrumentation
for Solar Astrophysics, ed. S. Fineschi & M. A. Gummin,
111–122
Xu, Y., Jing, J., Wang, S., & Wang, H. 2014, ApJ, 787, 7
Yang, K., Guo, Y., & Ding, M. D. 2015, ApJ, 806, 171
Zhang, Q. M., Ning, Z. J., Guo, Y., et al. 2015, ApJ, 805, 4
Zhu, C., Qiu, J., & Longcope, D. W. 2018, ApJ, 856, 27